A controlled experiment for evaluating a metric-based reading technique for requirements inspection

Natural language requirements documents are often verified by means of some reading technique. Some recommendations for defining a good reading technique point out that a concrete technique must not only be suitable for specific classes of defects, but also for a concrete notation in which requirements are written. Following this suggestion, we have proposed a metric-based reading (MBR) technique used for requirements inspections, whose main goal is to identify specific types of defects in use cases. The systematic approach of MBR is basically based on a set of rules as "if the metric value is too low (or high) the presence of defects of type de fType/sub 1/,...de fType/sub n/ must be checked". We hypothesised that if the reviewers know these rules, the inspection process is more effective and efficient, which means that the defects detection rate is higher and the number of defects identified per unit of time increases. But this hypotheses lacks validity if it is not empirically validated. For that reason the main goal is to describe a controlled experiment we carried out to ascertain if the usage of MBR really helps in the detection of defects in comparison with a simple checklist technique. The experiment result revealed that MBR reviewers were more effective at detecting defects than checklist reviewers, but they were not more efficient, because MBR reviewers took longer than checklist reviewers on average.